Time Series and Index Numbers for Forecasting

Learning objectives

• Explain the main components of a time series and why they matter for forecasting: trend, seasonal, cyclical and random variation.
• Calculate moving averages (including centred moving averages where needed) to smooth fluctuations and reveal underlying movement.
• Construct and interpret index numbers (simple and weighted) to compare changes over time and restate values on a consistent basis.
• Apply additive and multiplicative decomposition approaches to estimate trend and seasonality and produce forecasts.
• Evaluate forecast reliability by identifying common distortions such as outliers, structural change, shifting seasonality and base-period problems.

Overview & key concepts

Forecasting often starts with recognising patterns in historical data. Time series methods help separate long-run movement from within-year repetition and irregular effects. Index numbers complement this by expressing change relative to a base period, which helps comparisons across time (for example, converting cash figures into “constant price” terms so that movements reflect volume more than inflation).

Exam approach: Treat these questions as process-mark tasks. State whether you are using an additive or multiplicative structure (and why), then set out workings in a clear sequence (trend estimate → seasonal adjustment → forecast). Check your seasonal adjustments are correctly scaled (multiplicative indices average 1; additive factors net to zero). Finish with a brief comment on data limitations, one-off events and whether the forecast makes operational sense.

Time series components

A time series is data recorded in time order (for example, monthly demand, quarterly sales, weekly output).

The observed value at each time point can be thought of as being driven by:

• Trend (T):long-term direction (upward, downward, flat).
• Seasonal (S):repeating pattern within a year (quarters, months, weeks).
• Cyclical (C):multi-year expansions and contractions linked to the wider economy (often hard to measure cleanly).
• Random / irregular (R):one-off effects and “noise” that are not systematic.

In many exam-style datasets (especially short series), there is not enough information to estimate cyclical effects reliably. In practice, cyclical variation is often absorbed into the residual term, so models focus on trend + seasonality with remaining variation treated as unexplained noise.

Core theory and frameworks

1) Decomposition models

Two common structures are used:

• Additive model(seasonal effect is a fixed amount):
• Observed = T + S + R
• Multiplicative model(seasonal effect is proportional):
• Observed = T × S × R

Choosing the model:

• Useadditivewhere seasonal swings are broadly constant in size (e.g., +200 units each winter regardless of the underlying level).
• Usemultiplicativewhere seasonal swings grow as the series grows (e.g., winter is typically 20% above trend).

2) Moving averages and smoothing

Moving averages reduce short-term variation and make the underlying direction clearer.

For an n-period simple moving average:

Moving average at time t = (sum of n consecutive observations) / n

Key points:

• Larger n produces smoother results but reacts more slowly to changes.
• If n iseven(e.g., 4 quarters), the moving average falls between periods, so acentred moving average (CMA)is typically used.

How to centre (operational step):

1. Compute consecutive even-length moving averages (e.g., 4-quarter moving averages).
2. Centre them by averaging adjacent moving averages to align with an actual period.

For quarterly data:

Centred moving average for a quarter = (MA for periods 1–4 + MA for periods 2–5) / 2

This “average of two consecutive moving averages” is a common marking point because it places the trend estimate on a specific quarter.

3) Seasonal adjustment (multiplicative and additive)

Multiplicative seasonal indices

Once a trend estimate exists, multiplicative seasonal factors are found using ratios:

Seasonal ratio = Actual / Trend

Then:

1. Group ratios by season (e.g., all Q1 ratios together).
2. Average within each season to obtain seasonal indices.
3. Normalise so that the average index equals 1 (equivalently, indices sum to the number of seasons).

For quarterly data:

Normalised seasonal index = Raw seasonal index × (4 / sum of raw indices)

Additive seasonal factors

With an additive model, seasonality is measured as a difference (not a ratio):

Additive seasonal factor = Actual − Trend

Then:

1. Group differences by season and average within each season.
2. Normalise so that seasonal factors sum to zero over a full cycle.

For quarterly data:

Normalised additive seasonal factor = Raw factor − (sum of raw factors / 4)

4) Forecasting workflow (multiplicative model)

Method steps (use this sequence in workings):

1. Estimate trend (least squares or moving average/CMA).
2. Compute seasonal ratios: Actual / Trend.
3. Average ratios by season (Q1, Q2, Q3, Q4).
4. Normalise indices (sum to 4 for quarters; average to 1).
5. Project trend forward.
6. Forecast: Trend × Seasonal index.
7. Sense-check and comment on distortions/limitations.

5) Rounding discipline (practical convention)

Rounding can change results if applied too early. A common convention is:

• Keep trend values and ratios to at least 3 decimal places during calculations.
• Normalise seasonal indices before rounding.
• Present seasonal indices to 3 d.p. and forecasts to whole units (unless the question specifies otherwise).

6) Index numbers (simple and weighted)

Index numbers express relative change compared with a base period (often set to 100). They are widely used to compare prices, quantities or values, but the interpretation depends on what is being indexed.

Simple index (single item)

Simple price index = (Current price / Base price) × 100

Weighted indices (multiple items)

Weights reflect importance (often quantities or expenditure shares).

Two common constructions:

• Laspeyres (base-period weights):
• Laspeyres price index = × 100
• Paasche (current-period weights):
• Paasche price index = × 100

How to choose:

• A Laspeyres-style index holds the basket fixed and asks, “What would that same basket cost now?” It is practical when only base quantities are available, but it may exaggerate price growth if buyers substitute away from items that become relatively expensive.
• A Paasche-style index uses the current basket and can reflect current behaviour better, but it needs current quantities and can understate price pressure if substitution has already occurred.

Important distinction:

• Aprice indexmeasures price change, not volume change.
• Aquantity (volume) indexmeasures volume change, not price change.
• Avalue indexblends price and quantity effects. Mixing these interpretations is a common pitfall.

Deflation (restating values into constant base terms):

Real (constant base) value = Nominal value × (100 / Index)

Link to forecasting:
If inflation is material and you are modelling sales values (not volumes), consider deflating historic sales values before estimating trend. Otherwise, the “trend” may mainly reflect general price level increases rather than underlying demand.

7) Structural change (structural breaks)

If there is evidence that the pattern has changed (for example, a new store format, major pricing shift, product range change, or a change in trading hours), historic seasonality and trend may no longer be representative. Practical responses include:

• splitting the series into pre-change and post-change periods,
• re-basing indices around a more recent period, or
• using only post-change data if the old regime is no longer relevant.

Worked example

Narrative scenario

A retail company records quarterly sales volumes over two years to forecast demand for the next year. Management believes there is an upward underlying trend and a seasonal uplift in the fourth quarter.

Quarterly sales (units):

• Year 1 Q1: 480
• Year 1 Q2: 600
• Year 1 Q3: 520
• Year 1 Q4: 720
• Year 2 Q1: 520
• Year 2 Q2: 660
• Year 2 Q3: 580
• Year 2 Q4: 800

Required

1. Calculate the trend using a straight-line (least squares) method.
2. Determine seasonal factors using a multiplicative model.
3. Forecast quarterly sales for Year 3.
4. Identify potential distortions in the data.
5. Explain how the forecast is likely to affect key financial statement figures.

Solution

1) Trend (least squares straight line)

Let time period t run from 1 to 8 for the eight quarters given.

The least squares trend line is:

T = a + bt

Using the data, the fitted trend is:

T = 483.571 + 28.095t

Trend values for each quarter:

• t1: 511.667
• t2: 539.762
• t3: 567.857
• t4: 595.952
• t5: 624.048
• t6: 652.143
• t7: 680.238
• t8: 708.333

2) Seasonal factors (multiplicative)

Seasonal ratio each quarter:

Seasonal ratio = Actual / Trend

Ratios (3 d.p.):

• Year 1 Q1: 480 / 511.667 = 0.938
• Year 1 Q2: 600 / 539.762 = 1.112
• Year 1 Q3: 520 / 567.857 = 0.916
• Year 1 Q4: 720 / 595.952 = 1.208
• Year 2 Q1: 520 / 624.048 = 0.833
• Year 2 Q2: 660 / 652.143 = 1.012
• Year 2 Q3: 580 / 680.238 = 0.853
• Year 2 Q4: 800 / 708.333 = 1.129

Average ratios by quarter (raw seasonal indices):

• Q1: (0.938 + 0.833) / 2 = 0.886
• Q2: (1.112 + 1.012) / 2 = 1.062
• Q3: (0.916 + 0.853) / 2 = 0.884
• Q4: (1.208 + 1.129) / 2 = 1.169

Normalisation check (quarterly): indices should sum to 4 (average 1). The indices sum to approximately 4.000 (rounding differences), so no material adjustment is required.

Normalised seasonal indices (3 d.p.):

• Q1: 0.886
• Q2: 1.062
• Q3: 0.884
• Q4: 1.169

Data limitation note: two years of quarterly data provides only two observations per quarter, so these seasonal indices are relatively fragile and should be interpreted with caution.

3) Forecast Year 3 (trend projected and seasonality applied)

Year 3 corresponds to t = 9 to 12.

Projected trend:

• t9: 736.429
• t10: 764.524
• t11: 792.619
• t12: 820.714

Multiplicative forecast:

Forecast = Trend × Seasonal index

Forecasts (rounded to whole units):

• Year 3 Q1: 736.429 × 0.886 = 652
• Year 3 Q2: 764.524 × 1.062 = 812
• Year 3 Q3: 792.619 × 0.884 = 701
• Year 3 Q4: 820.714 × 1.169 = 959

4) Potential distortions in the data

The pattern shows a strong Q4 uplift in both years, but the uplift is stronger in Year 1 than Year 2 (Q4 ratio is higher in Year 1). Possible causes include:

• promotional intensity differing between years,
• stock constraints limiting sales in a quarter,
• significant changes in pricing, product mix or distribution channels,
• one-off events (weather disruption, local competition changes, unusual bulk orders).

With short datasets, a single unusual quarter can pull seasonal indices materially. Where possible, investigate and adjust for clearly non-recurring effects.

5) Likely effects on financial statement figures

A forecast is not an accounting entry, but it drives budgeting and operational decisions that affect reported numbers and estimates, for example:

• Revenue planning:supports sales budgets and performance targets.
• Inventory and cost of sales:higher forecast volumes usually mean higher inventory investment ahead of peak periods; this affects purchases, closing inventory and the timing of cost recognition through cost of sales.
• Working capital:growth often increases trade receivables (if sales are on credit) and may increase trade payables (if purchases rise), affecting liquidity and funding needs.
• Cash flow timing:seasonal peaks can create cash strain if inventory must be built before sales receipts are collected.
• Operating costs and capacity:staffing, warehousing, distribution and marketing often rise in peak quarters; budgets should reflect seasonality in costs as well as revenue.

Interpretation of the results

The forecast indicates a continuing upward trend with a pronounced fourth-quarter peak. Planning should focus on whether supply capacity, staffing and working capital funding can support the Year 3 Q4 uplift without stock-outs or service problems.

A final sense-check is essential: if there are physical or operational constraints (space, labour availability, supplier lead times), a purely statistical projection may exceed what is achievable unless planned investment removes the bottleneck.

Common pitfalls and misunderstandings

• Ignoring one-off events (promotions, supply disruption, exceptional contracts) when estimating seasonality.
• Mixing upseasonal(within-year repetition) andcyclical(multi-year swings).
• Forgetting to centre moving averages when the number of periods in the average is even.
• Normalising after rounding (normalise first, then round for presentation).
• Forgetting scaling rules: multiplicative indices average 1; additive seasonal factors net to zero.
• Assuming seasonality is stable when the business model or customer behaviour has changed.
• Using too little data and overstating confidence (short series produce fragile indices).
• Treating a price index as if it measures volume, or confusing value change with quantity change.
• Failing to consider structural breaks (major price change, product change, route-to-market change).
• Producing forecasts mechanically without checking operational realism.

Summary

Time series methods help separate long-run movement from repeating seasonal patterns and irregular effects. Trend can be estimated using least squares or moving averages (including centred moving averages when the averaging length is even). Seasonality can then be estimated using either multiplicative indices (ratios) or additive factors (differences), with correct normalisation essential for reliable forecasts.

Index numbers support consistent comparisons across time, especially when inflation distorts cash values. When modelling sales values, consider deflating historic values if price-level change is material; otherwise, trend may reflect inflation rather than underlying volume.

FAQ

What is the difference between additive and multiplicative models?

Additive models treat seasonality as a fixed amount added to (or subtracted from) the underlying level. Multiplicative models treat seasonality as a proportion of the underlying level. Choose the model that matches how the seasonal swing behaves as the series grows or shrinks.

How do moving averages help in practice?

Moving averages smooth fluctuations, making the underlying movement easier to see. They are especially useful when data is noisy or when a practical trend estimate is required without building a full regression trend line.

What is a centred moving average and how do you calculate it?

With an even-length moving average (such as 4 quarters), the moving average lies between periods. To centre it, average adjacent moving averages so the result aligns with a specific period.
Centred moving average = (MA1 + MA2) / 2

Why must seasonal adjustments be normalised?

Normalisation prevents seasonal adjustments from accidentally inflating or deflating the overall level of the forecast. For multiplicative models, indices should average 1 across seasons (sum to the number of seasons). For additive models, seasonal factors should sum to zero across a full cycle.

What can go wrong with index numbers?

Common issues include choosing an unrepresentative base period, using outdated weights, and confusing price indices with volume measures. Index interpretation must match what is being indexed (prices, quantities or values).

How do outliers affect time series forecasts?

Outliers distort trend and seasonality estimates, especially with short datasets. If an outlier reflects a one-off event, it may be better treated separately rather than built into seasonal indices.

When should you deflate data before forecasting?

If you are using sales values and inflation is significant, deflating historic values into constant-base terms helps ensure the trend reflects underlying activity rather than general price increases.

Summary (Recap)

This chapter explained how time series data can be separated into trend, seasonal, cyclical and random influences to support forecasting. It covered moving averages (including centred moving averages), least-squares trend estimation, seasonal indices under multiplicative and additive approaches (with correct normalisation), and the construction and use of index numbers. The worked example demonstrated a full multiplicative workflow, highlighted the limits of short datasets, and linked forecasting outputs to practical effects on inventory, working capital and cash flow.

Glossary

Time series
Data recorded in time order (for example, weekly output or quarterly sales) used to analyse patterns and project future values.

Trend (T)
The underlying long-run direction of a time series.

Seasonal variation (S)
A repeating within-year pattern (for example, quarter-by-quarter or month-by-month changes that recur).

Cyclical variation (C)
Multi-year movements often linked to the economic cycle; typically less regular and harder to isolate.

Random / irregular variation (R)
Unpredictable fluctuations caused by one-off events or noise.

Moving average
A smoothing method that averages a fixed number of consecutive observations to reduce short-term variation.

Centred moving average (CMA)
A moving average aligned to a specific period by averaging adjacent even-length moving averages.

Seasonal index (multiplicative)
A factor representing typical seasonal uplift or suppression relative to trend; indices are normalised to average 1.

Seasonal factor (additive)
A typical seasonal adjustment amount measured as a difference from trend; factors are normalised to sum to zero over a cycle.

Additive model
A decomposition approach where seasonality is treated as a fixed amount.
Observed = T + S + R

Multiplicative model
A decomposition approach where seasonality is treated as proportional.
Observed = T × S × R

Index number
A measure that expresses change relative to a base period (often base = 100).

Base period
The reference time point used for index comparisons; results depend on whether the base is representative.

Weighted index
An index that applies weights (such as quantities or expenditure shares) so that more important items have greater influence.

Deflation (restating)
Converting nominal values into constant-base terms using an index to remove price-level effects.